Mathematical Physics

JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 97, 417-440 (1983)

Eigenvalue Assignment of an Augmented Hyperbolic System by Linear Feedback

B. M. N. CLARKE

School of Mathematics and Physics, Macquarie University, N. Ryde 21 IS, Australia

Submitted by Kenneth L. Cooke

A class of augmented linear hyperbolic systems subject to a boundary control is studied. It is shown that the eigenvalues of the closed loop system subject to a feedback control, can be assigned arbitrarily, apart from an asymptotic condition on the distribution of the eigenvalues. The feedback vector is constructed in an appropriate Hilbert space. The results are applied to an important example of the synthesis of a distortionless transmission line system by linear feedback.

1. INTRODUCTION

In [ 1 ] we studied a class of linear hyperbolic systems which included scalar linear second order hyperbolic equations in one space variable. We considered the problem of assigning the system eigenvalues by means of a boundary control and linear state feedback. It was shown that, subject to an asymptotic condition, the countable set of closed loop eigenvalues could be specified by the choice of an appropriate “feedback.” Moreover the feedback was constructed. Russell [3] had previously studied the analagous problem for the case of distributed control, through a different approach and obtaining similar results.

In this paper we will consider an eigenvalue assignment problem for a class of systems which includes that studied in [ 11. That our extension is a significant one is supported by the comments of Fritz John from his rather nice discussion of the notion of well posedness (of the classical problems of partial differential equations).

While well-posed problems are eminently satisfactory to the mathematical physicist, they do not, however, always reflect realistically what is wanted. While they provide for effects from outside by means of boundary conditions, they neglect the back effects of the system on the outside world that must be present. [2]

We consider the situation of a linear hyperbolic system in one space variable, subject to a boundary control and coupled at its far boundary to a finite dimensional constant coefficient linear system. The augmenting of the linear hyperbolic system by a finite dimensional linear system describes the interaction of the distributed parameter system and the external environment.

411 0022.247)</83 $3.00

Copyright B 1983 by Academic Press, Inc. All rghts of reproduction in any form reserved.

418 B. M. N. CLARKE

A further comment of Fritz John from the same discussion is appropriate here.

Very often in practice the problem is not to find out what happens for every conceivable initial arrangement of a system, but how to arrange things to come out with a desired effect, or how to conclude as much as possible about the solution from pieces of information that are redundant, incomplete and only approximate.

121

We will show that subject only to an asymptotic condition, the eigenvalues of a class of closed loop augmented hyperbolic linear systems, can be arbitrarily assigned by a choice of feedback vector belonging to an appropriate Hilbert space.

The augmented hyperbolic system we consider is described by

(l-2)

(1.3)

(aI +P,,a, -Pdy(L t>=a*w(f), (1.4)

54 = Aw + By(1, t), (1.5)

w(O) = wo 3 (1.6)

where x E [0,1], t > 0, A(x) is a continuous 2 x 2 matrix, y. E L,[O, I], u E L,[O, oo), wo, a E G”, A is an n x n matrix, B an n X 2 matrix, and ao, al, PO, p, scalars. All functions and constants are taken to be complex in general. The control is applied at the boundary x = 0 and the coupling introduced at the boundary x = 1 which is the most extreme situation for a hyperbolic system due to the finite wave speed. We could in fact generalize further by including a coupled finite dimensional linear system at the boundary x = 0 but this only unnecessarily complicates the exposition without adding any further insight.

Moreover the more appealing problem is that of the control of a remote finite dimensional linear system by a control applied at the near boundary of a wave transmission medium.

Russell [4] has given an example of an augmented hyperbolic system where the control appears simultaneously in the differential equations (1. l), (1.5) and not in the boundary condition (1.3). Clearly this is a less exacting situation than the case we will consider. Moreover the system we study cannot be covered by the results of [4] which require an expansion of the solution (y(., t), w(t)) in terms of open loop eigenfunctions (U = 0), and such an expansion is available only when the closed loop system has a homogeneous boundary condition at x = 0.

AUGMENTED LINEAR HYPERBOLIC SYSTEMS 419

Russell has considered a quite general class of augmented (and deficient) hyperbolic systems. The treatment in [4] relies heavily on the passage to a scalar linear neutral equation of finite order by a process of eigenfunction expansion. We will develop an alternative vector functional equation in a natural way without recourse to an eigenfunction expansion or a detailed description of the distribution of eigenvalues.

2. THE EIGENVALUE PROBLEM

We denote the state of the system (l.l)-(1.6) by the pair (y(., t), w(f)) and define the state space Z = L,[O, I] x C”. With the inner product

((Y, w>; (z, v>);r= .i,i z*(x) Y(X) & + v*w (2.1)

the space SF is a Hilbert space. When u = 0 we say that )L E C is an open loop eigenvalue of the system

(1.1~(1.6) if

(~(4 4, w(Q) = eY+(x), G)

is a solution for some nontrivial (8, ti) E SY. It can be shown that the open loop eigenvalues consist of a countable set of complex numbers {L],j E Z) together with a set of M complex number {sj ; j = 1, 2,..., n). The corresponding open loop eigenfunctions, (qj, gj) are linearly independent and complete in Z and form a Riesz basis for Z. A Riesz basis has the property that there exists a unique dual biorthogonal basis. The open loop eigenvalues satisfy the condition

where

That is, the countable set {Lj} is asymptotic to the roots of

e21a - --Y (2.2)

which is the equation satisfied by the eigenvalues of the “base” case system obtained by setting u = 0, w(t) = 0, B = 0, A(x) = 0.

A well-tried technique for modifying the behaviour of linear systems is to

420 B. M. N. CLARKE

allow the control to be a linear function of the state. In our case we consider boundary control of the type

4f) = ((Y(*, 43 w(O); k7 f)>‘r (2.3)

for some (g, f) E 3. We speak of the system described by (1.1~(1.6), (2.3) as the closed loop system. If we now again seek solutions of the form

(Y(x,~ w(t)) = eP’W), w> (2.4)

for complex p and appropriate (4(x), w) E A?@, the values of p for which such a solution exists, we call the closed loop eigenvalues. Clearly in general the closed loop eigenvalues differ from the open loop ones and we can pose the question, what countable set of complex number (pj;j E Z} together with a finite set {uj; j = l,..., n) can be assigned as closed loop eigenvalues by appropriate choice of feedback (g, f) E R? In the sequel we will completely answer this question. Furthermore we will devise a procedure for constructing the required feedback.

On substituting (2.4) into (1. 1 )--( 1.6), (2.3) we find that p is a closed loop eigenvalue if and only if the following system of equations has a nontrivial solution (4(x), w) E R,

[ 1 ; -“1 $‘+A(x)4=P$,xE [O,$ (2.5)

(a0 + PO, a0 - PO) 4(O) = ((k 4; (fs fb,

(a1+P1,a,-Pl)4(l)=a*w, @I-A)w=Btj(I).

For the moment we denote A(x) o(x) = F(x). Then $(x) satisfies

(2.6)

(2.7)

(2.8)

Substituting into (2.6~(2.8) we obtain the algebraic system of equations,

ao+Po ao-PO 0

(a, + PI> ep’ (a, -/3,) eep’ -a*

-b, ep’ -b2e-p’ PI-A

/ ((4, w>; (f3; f>>z \

, -(al -/?,) ePp(‘-l)) F(t)

0 -e --PO--l) I

F(t) &

(2.9)


where B = [b,, b,] and bi are n-vectors. Assuming for the moment the nonsingularity of the matrix appearing on the left above, we recall the identity for partitioned matrices,

(P- QS-‘R)-’ -(P- QS-'R)-' QS-' 0 1, l[R' :j=la: ::I>

(2.10)

where we take

a,+P0 ao-Po 1, Q=[-i*], (a, +/3,)e”’ (a, +,>e-0’ R = I-b,&” -b,epp’], S=pI-A.

We compute

QS-'R = - ,'* [ 1 @Z-A)-‘(-b,e~“,-b,e~P’I

=[

0 0 a*($-A)-‘b,ep’ a*@l-A)m’b,e.-P’ I

and therefore

(P- QSPR)-'

= {a, +p, -a*($-A)-’ b,)eP’ [ ao+Po ao-PO

(a, -/?, -a*@Z-A)-‘b,}epp’ 1 -1

{a,-P1-a*@r-A)-'b,}e~"' -(cI~-/~~) 1 = -{aI +pI -a*(@--A)p’b,} ep’ a0 +Po I d@)’

where

We define the polynomials

po@) E de@1 -A),

P1@)-det b,a* PI-A----

aI +PI

(2.11)

(2.12)

422 B. M. N. CLARKE

p&J)-det i

b,a* pZ-A----

f-y, -P1

a,@) = a* Adj@Z-A)b,

“*+P1 ’

Then

a,@> = a* Adj@Z - A) b,

al-P1 .

(P - QS-‘R)-’

(2.13)

(2.14)

(2.15)

(a, -P,) I;$- 11 e-pJ a0 -&I =

I (a, +PJ &q- 11 eP( -(a, +P,)

I

P&) ep’ (ao-A&al +PJD@)’

where

(2.16

and

WI =Po@)(e”p - VI - e2’pa,@) f ya&). (2.17

Finally

(P-QS-‘R)-‘=&

From (2.18) we obtain

-(P- QS- %)-‘Qs-I=(&QS

ep’ =- D@)

I -

-‘RI-* 0 a*(@-Aj-’ I

a* Adj@Z-A)

al +P1 ya* Adj@Z-A) 1 * (2.

a1 -P,


Naturally the polynomials po, p,, p2, a,, a2 are not all independent and we find the relationship between them as follows. First we recall Schurs formula for the determinant of a partitioned matrix

= det P det(S - RP-‘Q) (2.20)

=detSdet(P-QS-‘R) (2.21)

whenever P or S is nonsingular. Directly from (2.9),

= (a0 + p,) det (aiizf?J”’ -‘* ] PI-A

- (a,, - &) det (*r-~~~~, @l -a*

PI-A 1 =(a,+Po)det@I-A)(a,-P,-a*@Z-A)~‘b,}e~”’

-(aO-b,,)det@Z--A){a,+/?,-a*@Z-AA)-’b,}eP’

= (a0 -/&>(a, +P,> e-“‘{(Y- e”“>P&) + e2’paAP) - Ya2@>L

(2.22)

where we used Schurs formula at the second stage. Alternatively applying Schurs lemma immediately,

= (a, +P,)(a, -/3,)e-“det PI--A -%I I

- (ao-/lo)(a, +p,)e“‘det @---A -$) I 1

(2.23)

Comparing (2.22), (2.23) we obtain the identities

PO@> - a*@> = P2@),

-Po@) + a,@) = -P,(P).

(2.24)

(2.25)

We use these identities to eliminate a,, a, in (2.17), (2.18) obtaining,

(P- QS-lR)-’ =&

424 B. M. N. CLARKE

and

W> = e”%@) - w2@>. (2.27)

The importance of the analytic function D@) will become apparent later. For the moment we merely remark that since p,@), pz@) are manic polynomials of degree n, the roots of D@) = 0 consist of a countable set of complex number {sj;j E Z } satisfying the condition

C le2”j - yI* < o3 jcL

together with n complex number {uj;j= I,..., n}. This can be seen by applying Rouche’s theorem.

Now applying identity (2.10) to the system (2.9) we obtain

[; ;](“p’)= [ P-Q;-‘R)-’ -(P-Q~;nT'Q~-']

i

((+Y w>; (Et, f)>x

I ' ((al +/I,) ep(‘-l) x 0 , -(a, -/I,) edp(‘-l)) F(r) d<

- i ; [bled-~), -b2e-P(‘-“] F(t) dc ,

Solving this equation for 4(O) using (2.26), (2.27), (2.19)

mP> O(O) = a0 +Po a1 + P, a1 +P,

e”“p,@) _ YP~@) e” yeP’a* Adj@Z - A)

a,-PO aI -0, aI -PI / /

YP2@) PO@) e*’ eO’a* Adj@I - A) -___

((a1 + PI) e“(‘-l), -(al -PI) epp(‘pb)) F(t) &

- -b,e-“(‘-“I F(l) dr

426 B. M.N. CLARKE

Using the identities (2.24), (2.25) and collecting terms,

We define the 2 x 2 matrix K(x, c, p) and 2 x 1 vector H(x, p) by

(e”“p,@) -D(p) H(x - <)} epcx-l)

a0 -PO =

-I

-( 1 ao+Po p ~)e-p~xtI-2~~ --y

ao-PO ' ( 1 "0 P2@) ep(x+" 3

{yp2@) + D@) H(x - 0) eep'"-" 1

_ YP,@) epx a0 +Po

p,@) ePwx)

a,-PO

(2.29)

(2.30)

and finally obtain the Fredholm integral equation of a type considered by Tamarkin [5],

~Q)~(X)=~~K(X,~,~)A(~)~(~)~S+H(X,P)((~~W)I (g9% (2.31)

which together with the equation

@I-A)w =Bt)(I) (2.32)

defines the eigenvalue-eigenfunction problem for the augmented hyperbolic system.


Henceforth all integrations are over an interval [0, I] unless otherwise is indicated. We make a few general comments concerning the eigenvalue problem. First, the open loop eigenvalues and eigenvectors are assumed known and satisfy the system of equations

(2.33)

(a0 + PO? a, -Po) W) = 0, (2.34)

(a, +P,,a, -P,)&)=a*k (2.35)

(dZ -A)& = B&). (2.36)

By comparing the open and closed loop system (2.5~(2.8), (2.33)-(2.36) we have

LEMMA 2.1. Let p be a closed loop eigenvalue with corresponding eigenfunction (4(x), w) then a necessary and suflcient condition that p be also an open loop eigenvalue is that

((43 w>; (Is 0) r = 0. (2.37)

Second, the eigenspace corresponding to an eigenvalue (open or closed loop) has dimension at most n + 1. This follows most easily from special case of (2.9) with A(x) = 0 in the open loop case.

Third, the assumption

is equivalent to D@) # 0 from (2.23). If D(p) = 0, the integral equation (2.31) is still valid as examination of the derivation will show that we needed only the adjoint of the matrix

P- QS-‘R

to obtain (2.31) and not its inverse. It is then easy to show that when D(p) = 0, (2.31) reduces to the necessary condition

((0, w); (g, fh = j ((a0 + PO) epo’, -(a, -PO) e”) 40 $63 &. (2.38)

Premultiplying the system (2.5) by ((a0 + /3,) epp.‘. -(a0 -PO) e’l) and integrating over 10, 11 leads to

428 B. M. N. CLARKE

0 = ((a0 + PO> e-O’, (a, - PO) 4 O(l) - (a0 + PO, a0 -PO> 40)

+ j ((a0 + PO> e+‘, -(a, - PO) ept) 44 W &

= ((a0 + PO) e+, (a0 -PO) ep9 O(O - (0th w>; (gy 0)~

+ j ((a, + PO> e-5 -(a, - PO) ep’) A (0 NO dt

Now if D@) = 0, (2.38) implies from above

((a0 + PO> em”, (a, -PO) e”‘> W = 0 (2.39)

which together with (2.7), (2.8) forms a system of algebraic equations for (4409 w>,

(a0 + PO) e-O’ (a, -PO) ep’ 0

aI +P1 aI -P, -a* (2.40)

-b, -b, pI - A

This system has a nontrivial solution (+(1), w) and 4(x) is then determined as the solution of the Volterra equation

The special case A(x) = 0 is of interest since the open loop eigenvalues are zeros of D@). The necessary condition (2.38) is then just the content of Lemma 2.1.

Returning to the question of multiple eigenvalues, (2.9) can have more than one solution (b(O), w) if and only if D(p) = 0. Then a necessary and suffkient condition for (2.9) to have a nontrivial solution is

((6 w>; (g, f)>r

- ((a, +/I,) ep+t) , -(a, -P,> epp(‘-r’)A(t) Cp(t) &

-B r‘[

eP(l-l) 0 0 -e-P”- 1) I

A (4 d(t) &I

0 = (yJ*, z*>

(2.42)


for all solutions of

P Q

o=(v*,z*) R s * [ I (2.43)

If p is an eigenvalue whose eigenspace is of dimension two, then (2.43) has two linear independent solutions. With no loss of generality we may choose w* = (rj?, , I$,) arbitrarily. Then

(~*,z*)=1//,(1,O,O)+y/,(O, l,a*@Z-A))‘)

and a pair of solutions of (2.43) are

(w*, z*) = (LO, 01, (w*, z*> = (O,P,@), a* Ad.iW -A)).

But substitution shows that neither satisfies (2.42) in general, leading to a contradiction. We conclude that the eigenspace corresponding to an open or closed loop eigenvalue has dimension one.

3. EIGENVALUE ASSIGNMENT

We pose the eigenvalue assignment problem as follows: given a set of distinct complex numbers .Y E (pi;j E Z ) U (c,,;j = 1, 2,..., n) satisfying

find a vector (g, f) EY such that the closed loop system consisting of (2.3 l), (2.32) has a nontrivial solution (4, w) E .X if and only if p E .7.

We have a representation of a distinguished solution of the Fredholm equation (2.3 1).

THEOREM 3.1. If p is not an open loop eigenvalue and D(p) # 0 then

40, P) = H(x, P) + \ G(x, c;, P> A (0 H(L P) d5 (3.2)

satisfies

D(P) $(x, P) = j” W, <, P> A(t) 445 P> & + D@> WC ~13

where G(x, 5, p) is the 2 x 2 matrix satisfying

(3.3)

D@) G(x, t, P) = j K(x, B, P) A (9) (3~. ti P> & + K(xt i$ P). (3.4)

430 B. M. N. CLARKE

Proof Fredholm equations analytic in a parameter, of the type (3.3) have been studied by various authors [ 5, 61 and the existence and uniqueness of G(x, <,p) follows immediately from the Fredholm alternative. We then have

J K(x> 6 P> A (0 442 P> &

=~W~,CP)A(~ H(~,P)+~G(~,II,P)A(~~)H(~,P)~~~ dt 1 i

ZZ Ji W, t-2 P> + 1 K(x, rl, P) A (~1 WV, 6 P> drl i A (4 WC P) dt

= N~)(d(x, P) - Wx, P>). 1

Comparing the equation satisfied by +(x,p), (3.3) with the canonical Fredholm equation (2.31) shows that 0(x, p) is a closed loop eigenfunction corresponding to an eigenvalue p, and normalized such that

(3.5)

When D@) = 0 and p is not an open loop eigenvalue the eigenfunction corresponding to p is given by (2.40), (2.41) and we have the necessary condition (2.38),

((+(-, P), 4~)); (g, f>>.;r = 1 ((a0 + PO> eppt3 -(a,, - PO> @‘I 40 443 &.

(3.6)

Finally if p is equal to an open loop eigenvalue with corresponding open loop eigenfunction (4, i;) we have ($(., p), w@)) = (4, &) and the necessary condition

(3.7)

Given a set ,Y’ satisfying (3.1), the eigenfunctions, which are well defined and can be determined explicitly by Theorem 3.1 and the subsequent discussion, form a Riesz basis for .P. As is well known there is a unique biorthogonal basis for CT, ((w(., pj), z(pj));j E Z} U ((tp(., a,), ~(0~)); j = 1, 2,..., n). We expand the feedback (g, f) in terms of the dual basis of&Y,

(3.8)


From the biorthogonality we obtain the coefftcients

ej = ((O(‘, Pj>Y w@j)>; (g, f))U9 jE z, (3.9)

Sj = ((bC.1 Oj>i w(aj>>i (g? f))r, j = 1, 2,..., n (3.10)

and by the relations (3.5t(3.7) each gj,g,! is determined. We confirm that the feedback constructed belongs to .z?‘. When only

finitely many pi belong to the set .P’ ’ = (P,~ E .P~; j E Z, DQj) = 0, pj is not an open loop eigenvalue}, we need only check the convergence in &X of the series

\‘ pie 7 7 ’

O@j> (WC.3 Pj), ‘@j)).

The fact that (e”Oi - r} E I, implies that (D@i); p,i E .;L - .Y ‘) E I, and hence that (g, f) E R’.

We need to consider the case when the set ,Y ’ contains countably many pj. Then the convergence of (3.8) in P’ is not as obvious since it is not at all clear that the coefficients { gj ; pj E .Y ‘} form an lZ sequence. We will show that this is the case presently.

Consider the open loop system corresponding to the case A(x) = 0, whose eigenfunctions (6, ti) corresponding to an eigenvalue ,u satisfy

(3.11)

(a, + PO 3 a0 - PO> 0) = 0,

(a, + PI3 aI 4) WI = a*+

@I-A)w=Bi$(l).

(3.12)

(3.13)

(3.14)

Since all the eigenvalues of the above system are zeros of D(L) = 0, we may replace the boundary condition (3.13) by (2.39), and the eigenfunctions (3, I%) are explicitly given by (2.40), (2.41) for p =pj, II = 0. Examination of the open loop system (3.1 l), (3.12), (2.39), (3.14) shows that the function q(x) is independent of i, in fact we may easily obtain

(3.15)

which satisfies (3.1 l), (3.12), (2.39). The collection (3(x, ,B~); Dbj) = 0)

432 B. M. N. CLARKE

forms an excessive set in L,[O, I]. If we delete n members such that the remaining correspond to roots of D(U) = 0 satisfying

we obtain a Riesz basis for L,[O, 11. The dual basis of L,[O, I] consists of the corresponding set {9(x, ,u~)}, where (3(x, pj); ;(Ju~)) E ,F are solutions of the adjoint system to (3.11)-(3.14),

1

a,+/& G*(o) 1

i i

= 0,

a,-&

(3.17)

+i*b=O, (3.18)

4w> * =-a, aI +PI

(3.19)

where

bE b1 b, ---. a, +P, aI -A

(3.20)

In a similar fashion to the previous case, we find that solutions of the adjoint system (3.16~(3.19) satisfy a condition

\ir*w =o (3.21)

which we may use to replace (3.18). Again $(x,,u) is then independent of Z(U) and we obtain

\ir*(x, P) = ((a, + P,) epux, -(a0 -PO> W. (3.22)


We now notice that the necessary condition (3.6) for p E .Y ’ to be a closed loop eigenvalue can be written

((44.3 PI, w@)>; (8, f>)r= (AC.1 H.1 PX ct.3 P)),.*,“.,,. (3.23)

In this case I+(., p) is given explicitly by (2.40), (2.41). The set ,Y“ is a subset of the set

! P.j;NFj)= 0, \’ (e2’uf-y12 < 03 ( JEP I

corresponding to the bi-orthogonal bases for L,[O, 11, {i$(x, ,ui)), ($(x, pi)} given by (3.15), (3.22). We now expand A(x) 4(x, p), p E .Y ’ in terms of the basis {db, Pi> 1

(3.25)

where using the biorthogonality,

cj@) = CA(*) +(‘3 P>, \ir(‘l P.j))L2[0.jl (3.26)

and

for positive real constants m < M. Clearly p E 9 implies p =,ui for some i E L and the question of

convergence of the expansion for (g, f) reduces to the question of the convergence of the sum

(3.27)

Now {w( ., ,u~); ,u~ E .!? ’ ) is a uniformly bounded set in L, [0, 11 and from the construction of the closed loop eigenfunctions {(4(x, ,u~), wO~,~)), ,uj E .Y ’ }, (2.40), (2.41) we can appropriately scale the $(x,,D~), ,uj E .Y ’ to ensure that

Combining the above remarks with the inequality (3.27) we obtain

(3.28)

(3.29)

434 B. M. N. CLARKE

and we have completed the proof that our construction (g, f) E R when S Y ’ contains countably many points.

It is interesting to note that since the eigenvalues of the open loop system corresponding to A(x) = 0, satisfy D(A) = 0, the above discussion shows that we can remove the perturbation of the spectrum caused by the introduction of A(x) # 0, by a linear feedback boundary control.

We have now completely solved the eigenvalue assignment problem for the class of augmented hyperbolic systems (1.1 t( 1.6). We collect our results so far.

THEOREM 3.2. Let .Y E (pj;jE Z} U {a,i;j= l,..., n} be a set of distinct complex numbers satisfying

Then there exists a (g, f) E .F’ such that the closed loop system (1. 1 ), (1.3), (1.4), (1.5), (2.3) has spectrum the set ,Y. Moreover (g, f) is given by (3.8), (3.9), (3.10), (3.5), (3.6), (3.7).

4. AN EXAMPLE

Consider an electrical transmission line of length 1 with resistance, inductance, capacitance and conductance per unit length R, L, C, and G respec- tively. Let a voltage source of internal resistance R, be located at x = 0 and a series network consisting of a resistance, inductance, and capacitance R, , L, , C, be located at x = I (see Fig. 1). The voltage and current at (x, t) E [0, I] x [0, co) is denoted by the pair (u(x, t), i(x, t)). Then

-=-L!!-Ri l3V

8X at ’ ai -&Ga, ax

(4.1)

(4.2)

40, t) = v&t) + R,i(O, t), (4.3)

v(Z,t)=R,i,(t)+L,~(t)+~!ri,(r)dr, (4.4 ) I 0

where the current in the series network is

iI = i(Z, t). (45)

AUGMENTED LINEAR HYPERBOLIC SYSTEMS

FIGURE 1

We define the state (y(x, t), w(t)) E,F by

w,(t) = i,(t), WI(f) = it i,(r) ds -0

(4-e)

and

(4.7)

We assume that L, C, R, G are constant on 0 < x < 1. Then from (4.4), (4.7)

(-) I

RI 1 -- WI L, L,C, ZZ W? 1 0

Performing a change of variable (4.7) we obtain the augmented hyperbolic system

LG+RC LG-RC (4.10)

436 B. M. N. CLARKE

CR, + \/LIc)Y,(O, t) - (&J - \/LIc)Y(O, 4 = vow -=u(t), (4.11) Jc

q, f> = - JcY,(L f> + &Y,(L f> = w,(f) = (LO) w(c), (4.12)

(*) [

R, 1

WI L, L,C, = w2 1 0 I

c

fiJL -- WI ) [ + L, L, w .(4.13)

? 0 0 I i ;I;;

2 3 :; 1

By a simple change of length scale we can assume, with no loss of generality, that fl= 1. We make the obvious identifications

ao+Po=Ro+~, ao-PO=-Ro+m,

LG-RC =constant LG+RC I

3

CZI+pl=-& a, +p, = &, a* = (l,O),

to obtain our standard form (l.l), (1.3)-(1.5). Consider the open loop system in the special case when LG = RC. It is

easy to show that in this case the solutions of (5.10) can be obtained from

(4.14)

where

6 = (LG + RC)/2 a and

satisfy the lossless transmission line

at- & az -

i

0

equations

0

I 3Z ax’ -- & We see from this that if LG = RC the wave shape on a transmission line is

preserved but the magnitude will be attenuated exponentially with distance.


In the parlance of transmission line theory, the line is said to be distor- tionfess. Clearly this is a highly desirable state of affairs. Unfortunately for a typical construction and typical materials the condition LG = RC lead to physically inconvenient values for one or other of the parameters L, C, R, G. In particular, the attentuation can be too large for the line to be of any prac- tical use. For these reasons, distortionless transmission lines are rarely met. It is interesting to note that development of digital communications was largely motivated by the need to overcome the cumulative distortion of signals. Regenerative repeaters are able to recognise the presence or absence of a digital pulse and regenerate the signal. The precise wave shape arising is thus of less significance than in the case of an analogue signal.

It is easy to show that for a distortionless transmission line, (~(4 fh w(t)> = e”(+( x , w is a solution of (4.10~(4.13) if and only if > > ($, w) # 0 satisfies

(ail + P” 3 a, - PO) 440) = 03 (4.17)

(a, +P,3a,-P,)W)=a*~, (4.18)

@I-A)w=Bl+(/), (4.19)

where we have assumed that fl= 1, 6 = $(LG + RC). Now the general solution of (4.16) is

e(Ll t b)(.X-0 4(x)= [ o

0 e-(P+6)(x-I)

I w

and substituting into (4.17)-(4.19) we obtain

i

(a0 +&) e-(o+s” (a, -/3,) e(D+6)’ 0 #I(0

aI +P, aI -P, -a* 42w = 0%

-h -b, PI-A Ii 1 W

The characteristic equation is easily obtained as

e2’(p+S)pI@) - yp2@) = 0

or

6(p) E e”“p,@) - j+~~@) = 0,

(4.20)

(4.21)

(4.22)

438 B. M. N. CLARKE

where

jT= ep2’“y. (4.23)

The case 6 = 0 obviously corresponds to a lossless transmission line. Now consider the case of a transmission line which is not distortionless

(LG # RC). Then by the results of the previous section, we can assign as closed loop eigenvalues, the roots of D@) = 0, by appropriate choice of (g, f) E A? The closed loop system (4.10~(4.13) and

44 = ((Y(*, t>, w(O); (g, f)), (4.24)

will have spectrum identical with that of an augmented hyperbolic system consisting of a lossless transmission line connected to the series LCR network. The important question is, does the closed loop system have any of the other properties of a lossless transmission line system? The answer to this question is emphatically, yes! In what follows we will replace the series LCR network by any linear passive network represented by a finite dimensional system. Consider the solution of the closed loop system in a series of closed loop eigenfunctions

where

“j = ((YoY WO); (w(‘T Pk)y ‘@,i>))~’

c; = ((YO, wOh (d-7 Ok), Z(ak)>),-,

and { (w(., pj), zQj)); j E Z} U { (w(., ok), ~(0~)); k = l,..., n} is the dual basis of P. Then

(Y(L 0, w(t)> = y CjePf'(9(L Pj>Y w@j>> jsz

+ 2 c;euk’($(k ok), w(ok))- (4.25)

Now since pj, j E Z, ak, k = l,..., n, are roots of D@) = 0, from (2.40), (4.21) we see that for each closed loop eigenvalue p, (4(&p), w@)) generates the same one-dimensional subspace of R2+” as for the case of p an open loop eigenvalue of an augmented system consisting of a lossless transmission line. Thus from (4.25), the signal at the “receiving” end x = I, of the closed loop


system, is identical with that appearing at the end x = 1 of an open loop system incorporating a lossless transmission line. From the viewpoint of an observer stationed at x = 1 and able to perform measurements of (y(l, t), w(t)), the closed loop system behaves exactly as an open loop system with R = G = 0.

If we consider the more general question of assigning as closed loop eigenvalues, the roots of 6@) = 0, S f O., corresponding to the open loop eigenvalues of a distortionless but not lossless system, we see that this cannot be achieved with a feedback control (4.24) since the hypothesis (3.1) is violated for .7 consisting of roots of &) = 0, 6 # 0. If we admit a boundary feedback control

u(t)= ky(0, I) + U"(t),

where k = (k,. k,) is chosen so that

t a, + PO - k, ao-PO-k2

(4.26)

or

k, k,e-‘Is ___- “0 +Po a,-PO

= 1 -e-Z’S.

then y is replaced by F= ye- *” The boundary condition at x = 0, then . becomes

where

a:,+P;,=a,+rR,,-k,, a;-/?A=u,,-Po-k2. (4.27)

Then apply feedback

to place the closed loop eigenvalues at the zeros of 6@) = e”Op,@) - $&@). Then from (2.40) we have that

(at, + PA) e-O’ (ah - /?A) eo’ 0

a1 +a, a, -A -a*

-b, -4 PI--A

440 B. M. N. CLARKE

where (0(x, p), w@)) is a closed loop eigenvalue corresponding to a root of fi@) = 0. Using (4.26), (4.27) in (4.28) we obtain

i

(a,+/?,)e-‘“+S” (a,-j3B,)e’P+S” 0 h(LP)

a,+/31 ' a1 -PI -a* $*(LP) =o

-b, -b, pZ-A Ii 1 w@)

which is identical to (4.21). As before we conclude that the closed loop eigenfunction at x = I, generates the same one-dimensional subspace of R * “I as that generated by an open loop eigenfunction corresponding to p, of a distortionless system with attenuation factor 6 # 0.

In summary, we have shown that in the case of an augmented hyperbolic system consisting of a transmission line connected to a voltage source at x = 0 and a finite dimensional passive linear network at x = I, there exists a feedback control

u(t) = WO, t) + ((yC.5 t>, w(t)>; k, f>> F

such that the corresponding closed loop system has precisely the same spectrum at that of an open loop augmented hyperbolic system consisting of a distortionless transmission line connected to the same linear network, with attenuation factor 6. If 6 = 0, then we can take k = 0 and the closed loop spectrum is identical to an open loop augmented hyperbolic system containing a lossless transmission line.

Moreover, from the standpoint of measurements of (~(1, t), w(t)), the closed loop system is identical to that of the corresponding distortionless open loop system.

REFERENCES

1. B. M. N. CLARKE AND D. WILLIAMSON. Control canonical forms and eigenvalue assignment by feedback for a class of linear hyperbolic systems. SIAM J. Conlrol Optim. 19 (1981), 711-729.

2. F. JOHN, Partial differential equations, in “Mathematics Applied to Physics,” (J. Roubine, Ed.), Springer-Verlag, Berlin/New York. 1970.

3. D. L. RUSSELL, Canonical forms and spectral determination for a class of hyperbolic distributed parameter control systems, J. Math. Anal. Appl. 62 (1978), 186-225.

4. D. L. RUSSELL, “Closed Loop Eigenvalue Specification for Infinite Dimensional Systems: Augmented and Deficient Hyperbolic Cases.” MRC Tech. Summ. Rep.. No. 2021, Univ. of Wisconsin, Madison, 1979.

5. J. D. TAMARKIN, On Fredholm integral equations whose kernels are analytic in a parameter, Ann. of Math. 28 (1927), 127-152.

6. F. TRICOMI, “Integral Equations,” Wiley-Interscience, New York, 1957.

Mathematical Physics

Documents

Transcript of Mathematical Physics